DCA: An MxN framework based on Parallel RMI
The availability of massively parallel computer systems has made possible the evolution of scientific simulations from serial, monolithic programs to high-performance, parallel applications. In this context, the "MxN problem" is concerned with communication between parallel components running on different number of processors.
The Remote Method Invocation paradigm can be an effective and efficient approach to furthering MxN technology for distributed parallel components. Part of this will require defining the semantics of parallel component intercommunication in a way that end users will find natural and meaningful.
The Distributed Component Arquitecture (DCA) is a CCA framework under development that features a novel approach to Parallel-Remote Method invocation based on SPMD collective calls rather than traditional RMI. This approach to parallel RMI aims to close the gap between direct-connected and distributed CCA frameworks.
This work supported by National Science Foundation Grants 0116050 and EIA-0202048, and Department of Energy's Office of Science SciDAC grants.
For more information, visit: http://www.cs.indiana.edu/~febertra/mxn